In early May, the Arch-I-Scan welcomed Dr. Sharon (Shuihua) Wang as the newest member of our research team. She is taking up the role of Postdoctoral Research Associate (Mathematics), and will be primarily tasked with the development, training and maintenance of the AI image-recognition and machine-learning service. As a means of further introduction, Victoria Szafara, the project’s research assistant, has interviewed Sharon:
VS: Welcome to the Arch-I-Scan team, Sharon! It’s great to have you aboard. Maybe a good place to start is to let you introduce yourself and give some details on your background?
SW: Thanks. It is my great honour to join the Arch-I-Scan team. My personal research directions include machine-learning, deep learning, pattern recognition and data analysis. I have focused on exploring the possibility of using machine-learning and deep learning for medical data analysis for many years. For example, I have worked with research teams which proposed deep learning methods for COVID detection from CT images. Over the past two years, I also worked with Professor Ivan Tyukin and Professor Sheila McCann (University of Leicester, School of Mathematics and Actuarial Science) on the use of deep learning for the scar tissue detection from CMR images.
VS: Wow! So I definitely see how your experience is very applicable to the machine-learning aspect of the Arch-I-Scan project. How do you think your work on Arch-I-Scan will be different or uniquely challenging compared to some of your past research projects?
SW: I have been working on medical image analysis for many years, but this will be my first-time working with archaeological data. What is nice is that this is still similar to some of my previous work in that, in both instances, I will be dealing with image data. For the medical images, the usual aims were to determine the obvious differences between normal and abnormal data. However, for Arch-I-Scan, which has aims to identify Roman pottery sherds of various typologies, we may face a different kind of challenge; patterns may not be so obviously distinguished with so many more categories than in the medical data analysis. Also, the raw data is coming from fragmentary pieces (sherds) rather than whole pottery vessels which may present an additional challenge.
VS: The other week, you spent some time on campus scanning pottery with Daan and me. What was that like for you?
SW: I was so excited to back to the campus to spend some time scanning pottery – especially the first part, since of course it’s been more than one year that I’ve been working from home. Finally, I was able to return to campus and meet my colleagues for real (as opposed to on screen). The campus is so different than from before the lockdown. I feel that we are all lucky to have survived the pandemic and it’s interesting to see how things have changed. Scanning pottery in the lab with you and Daan was also interesting since usually I just analyse the collected data; in the past, I’ve never had the chance to ‘meet’ the data subjects. This time, I was able to photograph the subjects myself – subjects in a field of study which is totally new to me.
VS: What kinds of things are you up to now in terms of working on the project?
SW: Actually, I’ve been doing some deep thinking about this project. After I scanned the data, it became clearer to me that there are so many different categories (vessel types) with which we are working, but the data for each category is limited so far. Also, there will be some other data-specific challenges: for example, the features and traits which distinguish pottery types from one another are not always so very different. Therefore, I’ve been gathering some basic ideas for this project, like how we may be able to use transfer learning, data augmentation, multiview data analysis and start with some easier tasks.
VS: What are you looking most forward to in this, your latest professional post?
SW: I am looking forward to seeing how well our proposed method works on the Roman pottery data, knowing that machine-learning has worked well for medical data analysis. I am also looking forward to possible concrete end-products; if the model works well, it may even be developed as an app which can be installed on a smartphone. With a workable AI service, identification of Roman ceramics may become easier for a greater number of people, who may not always be experts.
VS: Thanks so much, Sharon; it’s been great to hear more about you in your capacity as the Maths Postdoctoral Research Associate for Arch-I-Scan. Before we close this interview, I also want to ask just a little bit about you personally. Can you share any fun facts or details of things that you like to do when you’re not developing cutting-edge tech programmes?
SW: Actually, I had a very hard period during the lockdown as I worried so much about the future, the heavy housework and childcare. I had to find something fun to do at home, and so I decided to plant some vegetables and flowers in my garden. It really keeps me busy. I like to go to the garden centre to buy compost, seeds and pots. I have to divide my weekends into time slots – when to shop, when to plant. But it is really interesting, especially when you watch those plants three times a day and you can spot the differences, day by day.
Do you have your own question or questions for Sharon? If so, feel free to post in the comments section below!